Key Takeaways

  • Better models do not make enterprise data problems smaller, because an AI can retrieve from the right system and still answer wrongly when it lacks business definitions, relationships, ownership, quality information, and usage policies.
  • Actian’s Zeenea acquisition in 2024 became the foundation of the Actian Data Intelligence Platform, connecting technical metadata with business meaning across federated systems without moving the underlying data.
  • Governance has to be operational: executable data contracts, data products with clear ownership, data quality checks in integration pipelines, and AI-assisted stewardship keep policies enforced as data changes.
  • Sovereignty means control, enforcement, and proof over where AI workloads run, which model they use, and what data they can access, not just where data is stored.
  • The next challenge is keeping business context accurate and giving AI agents enough context to act usefully within clear governance boundaries.

There’s a tendency to assume that as AI models improve, enterprise data problems will somehow become less important. I think the opposite is happening.

A model can retrieve information from the right system and still give you the wrong answer. It might use an outdated definition of a customer, combine data that shouldn’t be compared, or overlook restrictions on how certain information can be used. The model may be functioning exactly as designed. The problem is everything it doesn’t know about the data.

This is where context becomes critical. Not context in the narrow sense of giving a model a longer prompt, but the business definitions, relationships, ownership, quality information, and policies that determine how enterprise data should be interpreted and used.

It’s also a useful lens for what we’ve been building at Actian.

When we acquired Zeenea in 2024, we saw an opportunity that went well beyond traditional data discovery. Its technology became the foundation for today’s Actian Data Intelligence Platform. Enterprise AI has exposed a weakness in data architectures that we’ve lived with for decades. We have become very good at storing, processing, and moving data, but far less effective at making its business meaning available to the systems that consume it.

For a human analyst, that gap is manageable. Experienced people know which customer definition to use, which metrics to trust, and whom to ask when something looks wrong. AI systems don’t have that institutional knowledge. Unless it is explicitly represented and made accessible, they are left to infer meaning from incomplete information.

That is becoming a fundamental architectural challenge as enterprises move from AI that answers questions to AI that takes actions.

It’s also central to how we’ve been evolving data intelligence at Actian over the past two years, making business context more accessible to the people and AI systems that depend on it.

Two years on, as more organizations have moved AI beyond experimentation and into production, the objective has become more important than ever: helping enterprises understand not only what data they have, but what it means and how to use it.

From Data Catalogs to Connected Business Context

Data catalogs have been an important part of enterprise data management for years, helping organizations discover information across increasingly complex environments. But finding a dataset is only the first step. Understanding what it means, how it relates to other data, and whether it fits a particular purpose is a different challenge.

Consider something as ordinary as a customer record. Sales might define an active customer as someone with an open opportunity. Finance might define it as an active billing relationship. Customer support might look at recent interactions. None of those definitions is necessarily wrong. They’re answering different questions.

If an analyst understands those differences, they can decide which definition applies. An AI system needs access to that same understanding, or it can produce a convincing answer based on the wrong assumptions.

This was part of the thinking behind our acquisition of Zeenea. Its approach to federated metadata management gave us a foundation for connecting technical information with business meaning, relationships, lineage, and ownership.

The federated aspect matters. Most enterprises aren’t going to consolidate all their data into one environment, and I don’t think they should. They have existing investments, regulatory obligations, different operating models, and good reasons for keeping certain workloads where they are.

What they need is a way to understand relationships between data assets without moving all the underlying data.

Our federated knowledge graph helps establish those connections. It connects metadata across systems, making it easier to understand how data assets relate to business concepts and to one another.

That doesn’t mean the graph magically resolves every conflicting definition. Someone still needs to establish which definitions are authoritative, who owns them, and how to apply them. But it gives organizations a much better foundation for doing that work.

Context is Only Useful if You Can Trust It

Another mistake I see organizations make is treating metadata, governance, and data quality as separate projects.

They have a catalog initiative, a governance program, a quality tool, and perhaps a team working on data products. Each may be doing useful work, but the connections between them are often weak.

In my O’Reilly report, Foundations for Better Enterprise Data Management, I described five foundations that need to work together: architecture, governance, data quality and observability, metadata, and people and culture.

Data observability is an important part of that picture. Understanding what a dataset means is useful, but organizations also need to know whether the data is reliable, how it changes over time, and when something goes wrong. Without that visibility, even well-documented data can lead to poor decisions.

That’s why we’re investing in data quality and monitoring earlier in the data lifecycle. With Actian DataConnect, we’re bringing AI-assisted data quality assessment, remediation, and continuous monitoring directly into integration pipelines, helping organizations identify and address issues before they affect downstream analytics and AI applications.

I deliberately included people and culture because technology doesn’t settle questions of ownership or accountability on its own.

You can document that a particular dataset is the authoritative source for a business metric. But what happens when the data producer changes its structure? Who is responsible for communicating that change? How do downstream consumers know whether the data still meets their requirements?

These are operational questions, not cataloging questions.

They’ve influenced how we’ve evolved the Actian Data Intelligence Platform. Rather than treating metadata as a static inventory, we’ve been investing in ways to connect it with data products, governance processes, and the people responsible for managing enterprise data.

The objective isn’t to create another layer of administration. It’s to make the information organizations already maintain useful in the decisions and workflows where it matters.

Making Data Governance Operational

I strongly advocate moving governance closer to where data is created and consumed, following a shift-left approach.

That doesn’t mean every governance decision should be decentralized. Large enterprises still need common standards and oversight. But a central governance team can only do so much if every change requires manual intervention.

Think about a data producer changing the format of a field used by several downstream applications. The change might be entirely reasonable from the producer’s perspective. For a consumer expecting the previous format, it could cause a failure or, worse, a result that looks plausible but is incorrect.

A governance document saying changes should be reviewed doesn’t prevent that.

Data contracts offer a more practical way to express expectations between producers and consumers. They can define requirements for structure, quality, and other agreed-upon data characteristics. When those contracts are executable, the relevant checks can become part of operational workflows rather than relying entirely on someone remembering to consult a policy.

Data products bring another important element: clearer ownership and accountability for information that others depend on.

That’s also why we’ve invested in capabilities to help organizations define and manage data products and data contracts, making it easier to translate governance requirements into operational practices.

Our research with BARC, Data Products and Data Contracts in 2026: The Foundation for AI Success, reflects the growing adoption of these approaches.

Of course, a contract is only as useful as the requirements it expresses and the processes supporting it. Organizations still have to decide what should happen when a check fails, who can approve exceptions, and how changes are communicated.

That’s where I think the real work of governance lies.

We’re also applying AI to some of the more labor-intensive parts of stewardship. Our Data Steward Agent is designed to assist with metadata and governance tasks, so people can focus on decisions that require expertise and judgment.

I don’t see automation as a substitute for accountability. If anything, the more we automate, the clearer accountability needs to become.

Making Context Available Where Decisions Happen

The next challenge is making all this understanding available to the people and applications that need it.

Conversational analytics is a good example.

Being able to ask a question in natural language is a significant improvement over requiring every business user to understand database schemas or write queries. But the interface is only one part of the problem.

Ask an AI system how many active customers you had last quarter, and it still needs to know what active means, which data is authoritative, how the relevant measures are calculated, and what the user is permitted to see.

Without that understanding, a more conversational interface can simply make incorrect answers easier to obtain.

I explored this in my whitepaper, AI isn’t the Problem. Your Data Model Is. The underlying data model matters because it captures relationships and business rules that a language model cannot reliably infer from column names alone.

We’ve been approaching this problem from several directions.

Actian AI Analyst brings governed conversational analytics to enterprise data, reducing the technical barriers between business users and the information they need. Its integration with the Actian Data Intelligence Platform allows organizations to use business definitions and relationships maintained in the catalog to inform how analytical questions are interpreted.

AI Analyst also uses defined metrics and relationships to guide query construction, rather than leaving those decisions entirely to a language model.

Our work on intelligent stewardship supports the metadata and governance foundations these experiences depend on.

These aren’t interchangeable capabilities. They solve different problems, and they don’t remove the need for sound data architecture. But they reflect the direction we’re taking: making enterprise data easier to understand and use without asking people to ignore the complexity underneath.

For customers, that can mean less time searching for information, less manual effort maintaining metadata, and a clearer path from a business question to an answer they can assess and act on.

Context Also Includes the Rules About Where AI Can Run

One area where this becomes particularly interesting is data and AI sovereignty.

Sovereignty is sometimes reduced to where data is stored. But an organization might keep sensitive data in a particular jurisdiction while still sending information to an external model or service as part of an AI workflow. Data residency alone doesn’t guarantee control over how that data is used.

In the O’Reilly report Data and AI Sovereignty: Understanding Sovereign Data and AI Across Cloud, On-Prem, and Edge, which I co-authored with Ole Olesen-Bagneux and Steffen Kläbe, we examine this broader challenge.

I think about sovereignty in terms of control, enforcement, and proof. Can you decide where a workload runs, which model it uses, and what data it can access? Can you enforce those decisions across the systems involved? And can you demonstrate that the controls actually worked?

Our BARC research, Where Enterprise AI Runs: Deliberate workload placement across cloud, on-premises, and edge, examines how organizations are making increasingly deliberate decisions about where to deploy AI workloads. I don’t see this as an argument against cloud. Different workloads have different requirements, and portability is rarely as straightforward as it looks on paper.

That’s part of the thinking behind investments such as self-hosted AI Analyst with bring-your-own-LLM support and VectorAI DB, including deployment in more resource-constrained environments. These capabilities provide additional architectural choices, although they don’t remove the operational responsibilities involved or solve every aspect of sovereignty.

Ultimately, these decisions depend on understanding the information’s sensitivity, its business purpose, the applicable policies, and the environment in which the system operates. That’s another dimension of the context enterprise AI needs.

Two Years in, What Have we Learned?

Two years on, the architectural vision that guided our acquisition of Zeenea remains central to how we’re evolving data intelligence at Actian.

That acquisition gave us a strong foundation in metadata management and federated discovery. We’ve built on that foundation by connecting metadata with governance, data products, intelligent stewardship, and new ways for people to interact with enterprise data.

One of the most important lessons has been that metadata management and governance can’t evolve independently. Understanding where data lives, what it means, and how it moves across systems is foundational. But organizations also need to turn that understanding into clear ownership, enforceable policies, and processes that can keep pace with change.

At the same time, enterprise AI demands have pushed us to think more broadly about deployment flexibility and control.

These are different technology challenges, but they share a common problem: organizations need to understand their data and the conditions under which they can use it.

Difficult questions remain. How do we keep business context accurate as underlying systems change? How do we make policies enforceable across increasingly distributed architectures? How do we give AI agents enough information to act usefully without granting them more freedom than they should have?

I don’t think a single model, catalog, or platform will answer those questions. They require architectural choices, clear ownership, and a willingness to connect disciplines that have historically operated separately.

That’s where I believe data intelligence is heading. Metadata needs to become more actionable. Governance needs to be more continuous. And context needs to be available not only to people discovering data, but increasingly to the AI systems using it.

Recognition Along the Journey

Two years after acquiring Zeenea, Actian was recognized for the first time in the November Gartner® Magic Quadrant™ for Data and Analytics Governance Platforms as the only new entrant, positioned as a Niche Player.

It’s a meaningful milestone in our journey. Our approach to governance has evolved from helping organizations discover and understand their data to making governance more operational. Through data contracts and AI-assisted stewardship, we’re embedding policies into data workflows, reducing manual effort, and helping organizations maintain control as AI adoption accelerates.

We’ve come a long way since then, although there’s still plenty to do. And I think the challenges we’re addressing today are even more relevant as enterprises move AI into production.

Actian has also received recognition in other independent analyst assessments. We were named a Major Player in the 2026 IDC MarketScape for Data Intelligence Platform Software, included in Constellation Research’s ShortList for Semantic and Context Management for Analytics and AI, recognized as a Leader and Ace Performer in QKS Group’s 2026 SPARK Matrix for Data Governance Solutions, and named an Exemplary provider in ISG Buyers Guides covering Data Intelligence, Data Products, and Data Governance.

These assessments use different methodologies and examine different parts of the market. I wouldn’t treat them as interchangeable, or as a substitute for what customers experience when using our technology.

What matters to me is what comes next. As AI systems become more autonomous, the challenge is making sure they have access to the right business context, operate within clear governance boundaries, and remain accountable to the people responsible for the data.

That’s the next phase of the work we started two years ago.


Gartner, Magic Quadrant for Data and Analytics Governance Platforms, Anurag Raj, Sarah Turkaly, Lulu Wang, Dalia Naguib, 12 November 2026.

Gartner and Magic Quadrant are trademarks of Gartner, Inc. and/or its affiliates.

Gartner does not endorse any company, vendor, product or service depicted in its publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner publications consist of the opinions of Gartner’s business and technology insights organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this publication, including any warranties of merchantability or fitness for a particular purpose.